Geometric representations for conceptual design using evolutionary algorithms

Shaw, David John 2006. Geometric representations for conceptual design using evolutionary algorithms. PhD Thesis, Cardiff University.

PDF - Accepted Post-Print Version Download (6MB)

Abstract

Civil engineering design problems are typically approached using traditional techniques i.e. deterministic algorithms, rather than via stochastic search such as evolutionary algorithms. However evolutionary algorithms are adept at exploring fragmented and complex search spaces, such as those found in design, but do require potential solutions to have a 'representation' amenable to evolutionary operators. Four canonical representations have been proposed including: strings (generally used for parameter based problems), voxels (shape discovery), trees and graphs (skeletal structures). Several authors have proposed design algorithms for the conceptual layout design of commercial office buildings but all are limited to buildings with rectangular floor plans. This thesis presents an evolutionary algorithm based methodology capable of representing buildings with orthogonal boundaries and atria by using a 3-section string with real encoding, which ensures the initialisation and evolutionary operations are not too disruptive on column alignments encoded via the genome. In order to handle orthogonal layouts polygon- partitioning techniques are used to decompose them into rectangular sections, which can be solved individually. However to prevent the layout becoming too discontinuous, an 'adjacency graph' is proposed which ensures column line continuity throughout the building. Dome geometric layout design is difficult, because every joint and member must be located on the external surface and not impinge on the internal void. This thesis describes a string-based representation capable of designing directly in 3D using surface area and enclosed volume as the major search parameters. The representation encodes support and joint positions, which are converted into a dome by constructing its corresponding convex hull. Once constructed the hull's edges become the structural members and its vertices the joints. This avoids many of the problems experienced by the previous approach, which suffers when restrictive constraints such as the requirement to maintain l/8th symmetry are removed. The aim of this thesis is to investigate how some civil engineering design problems, in particular structures, can be represented using evolutionary algorithms (EA) and contains two, independent experimental chapters on building layout design and geometric dome design (an introduction to EAs and design is also provided). Civil engineering design problems are typically approached using traditional techniques i.e. deterministic algorithms, rather than via stochastic search such as EAs. However EAs are adept at exploring fragmented and complex search spaces, such as those found in design, but do require potential solutions to have a 'representation' amenable to evolutionary operators. Four canonical representations have been proposed including: strings (generally used for parameter based problems), voxels (shape discovery), trees and graphs (skeletal structures). Several authors have proposed design algorithms for the conceptual layout design of commercial office buildings but all are limited to buildings with rectangular floor plans. This thesis presents an evolutionary algorithm based methodology capable of representing buildings with orthogonal boundaries and atria by using a 3-section string with real encoding, which ensures the initialisation and evolutionary operations are not too disruptive on column alignments encoded via the genome. In order to handle orthogonal layouts polygon- partitioning techniques are used to decompose them into rectangular sections, which can be solved individually. However to prevent the layout becoming too discontinuous, an 'adjacency graph' is proposed which ensures column line continuity throughout the building. Dome geometric layout design is difficult, because every joint and member must be located on the external surface and not impinge on the internal void. This thesis describes a string-based representation capable of designing directly in 3D using surface area and enclosed volume as the major search parameters. The representation encodes support and joint positions, which are converted into a dome by constructing its corresponding convex hull. Once constructed the hull's edges become the structural members and its vertices the joints. This avoids many of the problems experienced by the previous approach, which suffers th when restrictive constraints such as the requirement to maintain 1/8 symmetry are removed.

Item Type:	Thesis (PhD)
Status:	Unpublished
Schools:	Engineering
Subjects:	T Technology > TA Engineering (General). Civil engineering (General)
ISBN:	9781303207839
Date of First Compliant Deposit:	30 March 2016
Last Modified:	12 Feb 2016 23:12
URI:	https://orca.cardiff.ac.uk/id/eprint/54570

Citation Data

Actions (repository staff only)

Edit Item